Question about training process and "rounds"

[lambertwx]

I'm confused about your paper's use of the term "round" and your training scheme in general. Is "round" the same thing as your term t?

Am I correct in understanding that the training on a given round t looks like:

1. Given X, Y, C, and X_m, train the multi-label classification model (starting with its weights from round t-1) until it converges.
2. Generate the pseudo-masks for each image in X
3. Given the pseudo-masks and X, train the segmentation model (starting with its weights from round t-1) until it converges.
4. Generate the average masks c_i from the segmentation masks X_m. These will form set C for the next iteration.

[nyjshinibaba]

@lambertwx @dongzhang89 作者你们好，可能是没有@导致你们没有看到这条提问，我也有相同的问题，请问这个代码的循环部分是怎么跑起来的，我在代码中只看到了产生伪掩码和分割模型训练两个部分单独的，但是整个框架的运行没找到，您知道如何解决吗？？？？